Benchmarking a File-based Digital Library System Repository Architecture

Digital Library Systems (DLSes) have over the past few decades evolved into complex tools and services used to manage Digital Libraries (DLs). However, as the amount digitised and born digital content being generated increases, there is increasingly a growing need for much simpler tools for the storage, management and long term preservation of data. The simplification in the design of DLS components has obvious an implication of, among other things, adversely affecting overall performance of resulting tools and services. This paper builds on previous work, which resulted in a prototype simple repository design, by outlines experimental results from a series of performance benchmarks that were conducted on to determine the extent to which such a simple repository architecture would scale to provide acceptable response times. The designed experiments were executed on a dataset with 1,638,400 objects, and involved execution of common DL operations on linearly increasing workloads designed based on the simple architecture. In addition, the experimental setup was replicated on a commonly used DL software ---DSpace--- to provide comparative metrics. The results indicate that collection sizes with at most 25,600 objects yield desirable response times. Furthermore, performance degradation typically manifests in information discovery operations.